Table 3.

Effect of including RLN clusters in a CoxPh model for prediction of Overall Survival (OS) and Recurrence Free Survival (RFS) at 5 years, and Radiation Associated Dysphagia (RAD) at 6 months on the validation data. Clinical features are age, smoking status (never/current/former), and chemo (yes/no). (RM) radiomics clusters alone, (LN) Lymph node clusters, (RM + LN) combined 3-stage radiomics and lymph node clusters, (2F) a-priori radiomics features. The best and worst model in terms of validation AUC score are shown in green and red, respectively. Combined radiomics and LN clusters performed the best in all cases.

Model		Overall Survival (OS)				Recurrence Free Survival (RFS)				Dysphagia (RAD)
		AIC	Calib	Cind	AUC	AIC	Calib	Cind	AUC	AUC
Clinical + AJCC	None (Baseline)	696.75	14.34	0.64	0.61	797.55	19.47	0.62	0.61	0.73
	Additional covariate	ΔAIC	ΔCalib	ΔCind	ΔAUC	ΔAIC	ΔCalib	ΔCind	ΔAUC	ΔAUC
	+RM+LN	−4.016	−5.414	0.026	0.043	−2.921	32.189	0.041	0.052	0.012
	+RM+LN+2FA	−5.207	−0.617	0.037	0.060	−1.981	23.511	0.030	0.041	0.017
	+2F	−4.750	−6.946	0.019	0.034	−1.687	−1.572	−0.004	−0.002	0.005
		AIC	Calib	Cind	AUC	AIC	Calib	Cind	AUC	AUC
Clinical	None (Baseline)	705.07	9.91	0.61	0.60	799.42	30.07	0.54	0.55	0.65
	Additional covariate	ΔAIC	ΔCalib	ΔCind	ΔAUC	ΔAIC	ΔCalib	ΔCind	ΔAUC	ΔAUC
	+RM+LN	−9.385	−6.322	0.045	0.059	−6.126	−0.822	0.097	0.099	0.044
	+RM+LN+2FA	−14.386	−2.920	0.061	0.070	−5.398	−0.933	0.096	0.091	0.080
	+2F	−13.704	1.556	0.045	0.045	−4.219	6.262	0.049	0.040	0.058

^aAJCC is AJCC (8th edition).

^bClinical covariates used are age, smoking status (never/current/ former), chemo (yes/no).

^c2F includes tumor volume and intensity as predictive radiomic features directly into the model.